IBM Agent Identity launches to secure enterprise AI agents
IBM announced the general availability of IBM Agent Identity on October 27, 2026, at IBM TechXchange 2026 in Atlanta to give enterprises structured identity and access controls for autonomous software. The system registers agents in a centralized inventory, connects upstream human ownership to downstream actions, and issues short-lived, task-scoped credentials. For organizations running automated workflows across cloud platforms, this solves the challenge of unmanaged non-human access by replacing persistent administrative credentials with automated least-privilege permissions and audit trails.
How IBM Agent Identity tracks autonomous software in central inventories
The platform establishes an authoritative record for every active agent by capturing ownership, metadata, risk classifications, approved tool connections, and lifecycle status. To solve the problem of persistent credentials and unclear ownership, the system preserves delegation across identity providers and token exchanges so every action traces back to an originating human, approval state, or system. It issues OAuth-based, short-lived, task-scoped tokens with consent-aware authorization for sensitive tasks. Contextual authorization, policy enforcement, runtime risk evaluation, and credential control remain active while an agent runs, producing structured audit trails for compliance reporting.
How HashiCorp Vault 2.2 secures credentials for non-human workloads
The general availability of HashiCorp Vault 2.2 introduces dedicated secured accounts for software agents, automatically renewing their credentials and incorporating post-quantum encryption. The update adds vTPM support for Linux virtual machines, allowing workloads to authenticate using cryptographic proof tied to an enrolled virtual TPM rather than distributed secrets. When used alongside HashiCorp Vault Agent Identity, IBM Verify provides centralized identity, governance, and token exchange while Vault brokers just-in-time credentials for target systems that cannot consume delegated tokens directly. Vault shares context with the central record to enforce least-privilege access and enable real-time revocation if an agent attempts an unauthorized action.
How IBM watsonx Orchestrate AI agents are discovered and contained
Through an AI Gateway in tech preview, IBM watsonx Orchestrate AI agents, models, and Model Context Protocol servers can be discovered across third-party environments including Amazon Bedrock, Microsoft Foundry, and Google Gemini. A containment feature runs external agents in isolated sandboxes and provides administrators with a kill switch to immediately terminate any compromised agent. The catalog aligns with OWASP, CoSAI, and IBM AI Risk Atlas standards, surfacing performance metrics and EU AI Act compliance details for select agents. Built-in evaluation tooling, now generally available, lets teams test accuracy and quality before deployment while tracking model and token costs in production.
Why AI agent identity governance is crucial for modern enterprises
According to Gartner predictions from April 28, 2026, an average global Fortune 500 enterprise will deploy over 150,000 agents by 2028, up from fewer than 15 in 2025. However, only 13% of organizations currently report having effective AI agent identity governance in place. As software agents take on direct authority to execute transactions and access internal databases, uncoordinated credentials create severe operational risks. Integrating with NVIDIA OpenShell secure runtime software pairs protected execution environments with verified identities, giving enterprise security teams real-time control over autonomous workloads.
What it changes for enterprise security teams
For security and IT teams, the rollout eliminates hardcoded secrets and unmonitored agent permissions across hybrid cloud infrastructures. By replacing static administrative keys with short-lived, task-scoped tokens, enterprises reduce the risk of credential theft and lateral movement during automated tasks. Structured audit trails automatically link downstream machine actions directly to human supervisors, cutting the time required to complete compliance audits and investigate runtime security incidents.
Questions and answers
What is IBM Agent Identity and how does it work?
IBM Agent Identity is an enterprise security solution that creates a central inventory for AI agents, recording their ownership, risk levels, lifecycle state, and approved tool connections. It delegates human authority using short-lived, task-scoped tokens and consent-aware authorization, maintaining continuous risk evaluation and audit trails while agents run.
How does HashiCorp Vault manage AI agent credentials?
HashiCorp Vault manages agent credentials by brokering just-in-time, target-specific access for systems that cannot process delegated tokens. In Vault 2.2, agents receive secured accounts with automatic credential renewal, post-quantum encryption, and virtual TPM authentication, enabling immediate credential revocation if an unauthorized action occurs.
How can enterprises govern autonomous AI agents across platforms?
Enterprises govern autonomous agents by deploying single enforcement layers like the watsonx Orchestrate AI Gateway to discover third-party agents across environments like Amazon Bedrock and Google Gemini. They enforce policy controls through hardware-virtualized sandboxing, emergency kill switches, and integrated identity providers such as IBM Verify and Microsoft Entra.
MOZGII view
How we would build it
We would build an agent identity pipeline around a central inventory, short-lived access tokens, and container isolation. The central record would register every active agent, its human owner, approved tool connections, and risk level. When an agent requests access to a target database, an identity provider would issue a task-scoped token tied to the originating user. For legacy tools that cannot process delegated tokens, we would use a secret broker to issue temporary credentials. Security managers would establish authorization rules and handle manual approvals for sensitive actions.
This design would break if an agent runs inside a cloud platform without passing through the gateway, bypassing audit logging. We would track the share of downstream agent actions linked to an active human supervisor as the main metric. Next, we would implement runtime policy enforcement that revokes credentials instantly if an agent strays from its assigned workflow.
Sources
Where the facts in this story come from.